The Prediction Agent Economy

The first autonomous agent economy is here, and it’s a great Olas showcase

I often get asked things like, "what will the killer use cases for Olas be?" and "where are all those transactions coming from on Gnosis Chain?" They're fair questions, and there really aren’t adequate resources answering them yet – that’s the aim of this post.

The Story So Far

We started in earnest building the system during the middle of last year. My co-founder David Minarsch met Martin, co-founder of Gnosis at Zuzalu. Together they said "prediction markets so far have been held back by the cost of human participation – what if we had agents do the work instead?" The idea was compelling, so we immediately started work on it.

Fast forward to today and Olas agents, participating in prediction markets, are responsible for over 300k transactions on Gnosis Chain. That's over 10% of all Safe transactions there for the duration of the chains whole life.

Furthermore, the system was recently cited in Vitalik's AI x Crypto article.

How does it work today?

This system works by coordinating agents around prediction markets to work together, specialise and creating a thriving economy.

I've truly never seen anything quite like it, so I'm excited to show you how it works.

Meet the Agents

A diverse gang
A diverse gang

Market Creators These guys listen for potential market opportunies, process those requests and deploy prediction markets on the Omen protocol. They also seed the market with liquidity. Note that thus far, they don't really have a profit motive and are more or less hardcoded to create new markets based on current events.

Traders These guys watch for new markets being created on Omen by Market Creators. When they see one, they spring into action, and pay Researchers to go off and help them research the question, and come up with a probability and confidence score which they can use to make bets in the prediction market.

Researchers (Mechs) This is potentially the most interesting of all the agents. Technically, Market Creators, Traders and Researchers don't have direct access to intelligence sources themselves. That is, they're not talking to some Bittensor model, or directly pulling onchain data. These agents outsource their intelligence to Researchers, making them highly adaptable and extensible.

Note in Olas-land these are somewhat confusingly called "Mechs".

Closers When the prediction market gets to the point where the answer can be known – e.g. when the presidential debate has actually finished and the result is public – Closer agents go off, find that answer and use it to definitively close the market.

How the agents interact

  1. Market Creators deploy markets and add liquidity. In the long-run they do this because making markets can be profitable.

  2. Traders see these new markets.

  3. Traders request predictions from Researchers (aka Mechs). They learn over time which Researchers provide the best predictions. Researchers use various AI models, logic and data sources to come up with their predictions. They sell them back to Traders.

  4. Traders take these predictions and use them to place bets on the prediction market.

  5. The overall activity of these Traders gives external parties "the prediction". If 90% of Traders are saying "yes" to a particular question, that is a strong signal that the event will happen.

  6. Finally, prediction markets need to be 'decided', which means the final answer needs to be provided. When the prediction market is coming up to its closing point, i.e. the point where the final answer should be known, Closers pay Researchers to find the real answer.

Here's another, more linear way to look at it:

The system has been trained on current events questions, like: "Who's going to win the presidential debate on Friday night?"

Where are all the transactions coming from?

There are hundreds of thousands of transactions being made by these agents. But why are they actually making transactions? It's a great question, and illustrates a good reason why agents in crypto, and AI more broadly, makes sense.

  • Market Creators create transactions to deploy new prediction markets onchain. They may also update them.

  • Traders make transactions to place bets on those prediction markets. They also pay Researchers to come up with probabilities for them.

  • Closers make transactions to pay Researchers to find markets' final answers. They also need to issue transactions to actually close the markets.

  • Note also that each of these agents are registered and configured via the Olas protocol – all of these actions also require onchain transactions to be made.

Why are people running these agents?

There are two main reasons:

  1. Staking incentives

  2. Profit opportunity

Staking Incentives Valory has an ongoing program that incentivises people to run Traders. It's called Staker Expeditions and is based on the PoAA mechanism.

Staking here essentially says – any agent who stakes enough OLAS and performs a certain type of activity everyday will be eligible to earn rewards. This is a fantastic mechanism for getting the agents participating and improving, particularly before the agents have matured to be reliably profitable.

Currently, only Traders are incentivised through staking rewards. At some point, it could make sense to incentivise the other agents, like the Market Creators and Closers, to get them discovering market opportunities and reliably closing markets.

Profit Opportunity The other main reason for people to run these agents, of course, is that there is potential for them to be directly profitable. If you run an agent that reliably makes winning bets, then it is feasible for it to reliably earn money for its owner.

Who cares about predictions though?

This isn't some sort of academic experiment. From a product perspective, the possibility of generic predictions-as-a-service represents an entirely new product category, and something to be extremely excited about.

Whilst current events were a logical place to start training the system, I personally don't find them that interesting.

What is extremely exciting, however, is the potential of this system to become a generalised, low-cost predictions-as-a-service product. What could it be trained to make predictions about?

  • "What's the likelihood of this DAO proposal improving the DAO's annual recurring revenue by more than 50%?"

  • "What's the likelihood that this smart contract gets hacked or rugged?"

  • "What's the likelihood that this date will be successful?"

  • "What's the likelihood that I'll enjoy watching this film?"

Currently these types of questions could probably be answered if you had a ton of data, your own machine learning team and an entire company to throw at the question. But virtually no-one will ever have that.

Olas' prediction economy points towards a world where asking for predictions about any question becomes possible.

Furthermore, it points to a world where agents are requesting predictions from one another to guide their processes. Where agents decide whether to vote for a DAO proposal, or to ape into memecoin. Predictions exchange will be a fundamental primitive of the agent economy – thousands or even millions delivered per second.

Where should we go from here?

So it turns out that us crypto degenerates have developed a very plausible first autonomous agent economy. Well done to us. But where do we go from here?

Expand the Prediction Economy

When describing this system in person, we often get questions that try to tear apart the system. Let me be clear – it's not perfect from every angle. It's not perfectly accurate yet, and it isn't fully autonomous. There's so much that can be done to make this economy better:

  • Traders can have better strategies – what's the optimal way to bet on a prediction market? Which Researchers provide the best predictions?

  • Market Creators can learn to better recognise market opportunies – which types of markets will generate the biggest trading fees?

  • Researchers can get waaaay better – can we build better fine-tuned models and datasets for making predictions? Can the way Researchers use those models be refined?

So many of you out there are incredibly smart, and Olas is designed to reward you for your efforts. Check out the section "How can you get involved?" below to understand how you can contribute.

Expand the Agent Economy Universe

It's important to note that Olas is not only concerned with building a single economy focused on predictions. Olas is concerned with how to enable a rich universe of interdependent autonomous agent economies.

One of the most important things we learned definitively when contributing to this prediction economy is that control over the incentives – the incentives for people develop and run the agents – is the secret sauce for building sustainable and valuable agent economies. This is what led to the initial Proof of Active Agent design.

Subsequently, Valory has made this much more general, and made it so that Olas can use it to generate entirely new economies, focused on whatever it wants, powered by the OLAS token. This is what is at the heart of the recent Olas Staking announcement which you can learn more about here:

How do we take this prediction economy and make it so that Olas can spawn loads of economies, focused on anything it wants?

-> That's Olas Staking

How can you get involved?

There are loads of ways. A quick list:

  • Build "Mechs tools". These are basically Researchers, as I've framed it in this article. If you build these you can get rewarded by the Olas protocol.

  • Improve Trader strategies. Again, this can earn you rewards.

  • Run Trader agents.

  • If you find what we've built here exciting, tell people about it. Let me know if you want to learn more and want further resources to learn from – we need feedback.

  • Integrate your infrastructure products. I didn't cover it here, but these agents are consuming infra left, right and centre – Nevermined for payments, Nodies for RPCs, various AI models – we even tried to integrate Bittensor, alas without luck. Reach out to me if you're interested to learn how you could integrate your product and start attracting agent users.

  • Build products on top – if you have a product that could use these predictions to power it, that can be worked out. Again, reach out.

More generally, if you've enjoyed these ideas and are interested to get involved with making the universe of agent economies a reality, do jump in. Send me a message on Twitter or Telegram and I'll help you get started.

Subscribe to Oak
Receive the latest updates directly to your inbox.
Mint this entry as an NFT to add it to your collection.
This entry has been permanently stored onchain and signed by its creator.